My research program emphasizes in the exploration, development and implementation of control and estimation methods to address real world problems via provably correct solutions. I am particularly interested in complex multi-agent systems operating in uncertain environments (e.g., unstructured and/or adversarial).
Some of the ongoing and past research programs are described below.
A Human-centric Network of Free-Flying Co-Robots: We envision to ultimately augment astronauts in spacewalk with the Astronet: A Network of Astronautical Free-Flying Co-Robots, such as SPHERES or Astrobees developed by NASA. The Astronet will safely surround the crew member during EVAs, infer human intent and interpret them into predefined tasks, and respond to human inputs by redistributing autonomously in space to dynamically and continually improve task conditions in a human-centric way. We link our dynamic coverage control paradigm for the swarm motion with inference methods to online learn the task-relevant cues of the human while executing multiple tasks. We also motion models and sensing uncertainty for the 3D flight of free floating robots, and energy consumption models capturing their limited power resources.
This research is sponsored by the NASA Space Technology Research Grants Program through an Early Career Faculty Award "https://www.nasa.gov/directorates/spacetech/strg/ecf2016/AstroNet.html"
W. Bentz, S. Dhanjal and D. Panagou "Unsupervised Learning of Assistive Camera Views by an Aerial Co-robot in Augmented Reality Multitasking Environments", Int. Conf. on Robotics and Automation, Montreal, Canada, May 2019.
W. Bentz and D. Panagou "Bayesian-inferred Flexible Path Generation in Human-Robot Collaborative Networks", Int. Conf. on Robots and Intelligent Systems, Madrid, Spain, October 2018.
On-the-Fly Assistive-View Learning in Augmented Reality Multitasking Environments
Human-Robot (Astrobee) Interaction in a Virtual Reality Environment of the International Space Station
Human-Robot Interaction for Dynamic Coverage via Gesture Following
Bayesian-inferred Human Intention and Flexible Robot Trajectory Generation
The goal of this research project is to narrow the existing gap between high-level discrete task planning and low-level continuous control in complex multi-agent missions. We develop the concept of Multiple Finite-Time Control Barrier Functions to enable the definition of consistent mappings between high-level specifications and low-level safety controllers. Finite-Time Control Barrier Functions capture spatiotemporal specifications and interactions among agents, e.g., safety and time constraints. As of now we have extended classical results on the stability of switched and hybrid systems to Multiple Finite-Time Lyapunov Functions that enable convergence of system trajectories in finite time. Currently we are working on enabling prescribed-time convergence for the Multiple Control Barrier Functions to address explicit time constraints as well.
This research is sponsored by the Air Force Office of Scientific Research through an Young Investigator Award "http://www.wpafb.af.mil/News/Article-Display/Article/969772/afosr-awards-grants-to-58-scientists-and-engineers-through-its-young-investigat/"
K. Garg and D. Panagou "New Results on Finite-Time Stability: Geometric Conditions and Finite-Time Controllers", 2018 American Control Conference, Milwaukee, Wisconsin, June 2018
We consider resilience AND safety of multi-agent networks (e.g., multi-vehicle systems) in adversarial environments. Safety is viewed as the guaranteed collision-free motion of the agents while collaborating towards a common task (e.g., data gathering). Collaboration in principle requires coordination and negotiation mechanisms among the agents; these mechanisms are implemented using information shared over wireless communication links. However, wireless communication is vulnerable to cyber-attacks.
Resilience is hence viewed as the guaranteed accomplishment of the mission, despite the presence of possible adversaries that can send malicious data over compromised communication links. Our goal is to establish resilient communication structures, as well as estimation (filtering) and control mechanisms that will allow the multi-agent network to tolerate or mitigate the adversarial effects of malicious data in the network, while still maintaining safety guarantees.
Our premilinary work includes the establishment of k-circulant graphs as a sufficient communication topology for achieving resilient asymptotic consensus, as well as the extension of strong r-robustness graphs to achieve consensus to arbitrary reference values, that can be used in Leader-Follower networks. We are currently working on incorporating finite-time detection of adversaries and Control Barrier Functions for guaranteed safety despite the destabilizing effects of adversarial agents in the network.
This research is sponsored by Automotive Research Center and US Army TARDEC under the project "Adversarially Robust Coordination for Autonomous Multi-Vehicle Systems", and by the Army Research Office (ARO) under Award No W911NF-17-1-0526.
J. Usevitch, K. Garg and D. Panagou "Finite-Time Resilient Formation Control with Bounded Inputs", 57th IEEE Conference on Decision on Control, Miami, FL, December 2018
J. Usevitch and D. Panagou "Resilient Leader-Follower Consensus to Arbitrary Reference Values", 2018 American Control Conference, Milwaukee, Wisconsin, June 2018.
J. Usevitch and D. Panagou "r-Robustness and (r,s)-Robustness of Circulant Graphs", 56th IEEE Conf. on Decision and Control, Melbourne, Australia, December 2017.
Dynamic coverage is defined as forcing an agent to sense/cover over time each point of a domain of interest up to a satisfactory level. For agents with sensing functionals defined over finite footprints, this formulation results in algorithms which set them in motion based on how well they sense the surrounding environment. Thus agents are forced to autonomously and continually explore an unknown region (search) so that each point of this region is sensed for a prescribed amount of time. This requirement is encoded in a coverage metric that expresses the quality of information accumulated over time through the agent’s sensing footprint. The dynamic coverage problem then reduces to deriving the control laws for the motion of the agents so that the associated coverage error is driven to zero. These control laws force the agents to autonomously move towards, and consequently explore, non-searched regions.
We have developed energy-aware decentralized control algorithms for the motion of agents performing coverage tasks, along with collision avoidance guarantees under anisotropic sensing that creates asymmetric interactions among agents.
D. Panagou, D. M. Stipanovic and P. G. Voulgaris "Distributed dynamic coverage and avoidance control under anisotropic sensing", IEEE Transactions on Control of Network Systems, vol. 4, no. 4, pp. 850-862, 2017.
Aerial Sensing Networks in 3D Environments: We have extended our dynamic coverage control approach to the problem of increased and sustained 360 Situational Awareness in 3D environments. We developed an Aerial Sensing Network of small UAVs, which are actively exploring the area around a Ground Station by means of Energy-Aware Dynamic Coverage. Planning and control are tailored to the 3D motion constraints of small UAVs, the limitations of data links, cameras and other onboard sensors, as well as on the UAVs' energy (battery life) constraints. 3D coverage metrics capturing visual information gathering and navigation in 3D spaces yield control algorithms that ensure effective 3D visual coverage. The remaining battery life of each agent is taken into account so that power-constrained agents are redistributed in space; this allows those agents with longer battery life to explore farther away from the Ground Station, and those with shorter battery life to return safely to the Ground Station. Collision-free trajectories are generated in real-time through novel 3D coordination and collision avoidance algorithms with provable guarantees.
This research was sponsored by the Automotive Research Center and US Army TARDEC under the project "SQUAD: Situational Awareness and Sustained Survivability through Man/Unmanned Teaming" "http://arc.engin.umich.edu/research/5_A46_Situational_Awareness.html"
W. Bentz and D. Panagou "A Hybrid Approach to Persistent Coverage in Stochastic Environments", under review
W. Bentz, T. Hoang, E. Bayasgalan and D. Panagou "Complete 3-D Dynamic Coverage in Energy-constrained Multi-UAV Sensor Networks", Autonomous Robots, 2018.
Outdoors Flight over the University of Michigan Wave Field
We develop distributed coordination algorithms for multi-agent systems while respecting certain safety and performance guarantees. Safety is realized as the collision-free navigation of the agents towards goal locations under restricted sensing and communication capabilities. Performance is primarily realized as the robustness of the derived solutions against communication/sensing uncertainty and malicious behavior by the agents.
We address multi-task problems (such as collision avoidance, connectivity maintenance and convergence to goal destinations) for networks of mobile agents via a novel class of Parametric Lyapunov-like Barrier Functions. The Parametric Lyapunov-Barrier Functions capture safety and convergence specifications, as well as sensing and communication misinformation among the agents, and ensure robust connectivity and safety for the agents against the considered disturbances. Our next goal is to extend the methodology to realistic models for a wide class of autonomous vehicles.
We also consider coordination for agents belonging to different classes, namely class-A and class-B. Agents of class-B do not share information with agents of class-A and do not participate in ensuring safety, modeling thus agents with failed sensing/communication systems, agents of higher priority, or moving obstacles with known upper bounded velocity. We propose the notion of semi-cooperative coordination: Semi-cooperative coordination is defined as the ad-hoc prioritization among agents of the same class; more specifically, participation in conflict resolution and collision avoidance for each agent is determined on-the-fly based on whether the agent's motion results in decreasing its distance with respect to its neighbor agents; based on this condition, the agent decides to either ignore its neighbors, or adjust its velocity and avoid the neighbor agent with respect to which the rate of decrease of the pairwise inter-agent distance is maximal. Guarantees on safety and almost global convergence of the agents to their destinations are formally proved.
Agents of Class-A and Class-B (in V-shape formation).
D. Han and D. Panagou "Robust Multi-task Formation Control via Parametric Lyapunov-like Barrier Functions", IEEE Transactions on Automatic Control, Conditionally Accepted, June 2018.
D. Panagou "A Distributed Feedback Motion Planning Protocol for Multiple Unicycle Agents of Different Classes", IEEE Transactions on Automatic Control, vol. 62, no. 3, pp. 1178-1193, March 2017.
D. Panagou, D. M. Stipanovic and P. G. Voulgaris "Distributed coordination control for multi-robot networks using Lyapunov-like barrier functions", IEEE Transactions on Automatic Control, vol. 61, no. 3, pp. 617-632, March 2016.
Deconfliction in Multi-sUAS Missions: We considered multi-sUAS missions with heterogeneous aircraft (fixed-wing and multi-copters). Our focus was on designing computationally efficient mechanisms that deconflict multiple agents in the presence of sensing uncertainty, wind, and physical sobstacles. The proposed work will enable fast and accurate trajectory generation for self-separation in complex environments.
This research was supported by the NASA Grant NNX16AH81A.
K. Garg and D. Panagou "Hybrid Planning and Control for Multiple Fixed-Wing Aircraft under Input Constraints", Best Student Paper Finalist, 2019 AIAA Science and Technology Forum and Exposition (Scitech) Forum, San Diego, CA, January 2019
K. Garg, D. Han and D. Panagou "Robust Semi-Cooperative Multi-Agent Coordination in the Presence of Stochastic Disturbances", 56th IEEE Conf. on Decision and Control, Melbourne, Australia, December 2017.
X. Ma, Z. Jiao, Z. Wang and D. Panagou "3D Decentralized Prioritized Motion Planning and Coordination for High-Density Operations of Micro Aerial Vehicles", IEEE Transactions on Control Systems Technology, May 2017, DOI: 10.1109/TCST.2017.2699165.
Aircraft (Fixed-Wing) Deconfliction via Hybrid Control.
Decentralized Goal Assignment and Trajectory Generation via Multiple Lyapunov Functions | Collaborative work with UPenn: We develop decentralized feedback control policies and coordination protocols for multi-robot systems with certain safety and performance guarantees. Safety is realized as the collision-free motion towards goal locations under restricted sensing and communication capabilities, while performance is realized as the assignment of goals which result in shortest total distance to the goal locations. The formulation within a Multiple Lyapunov-like Barrier Functions approach enables scalable and correct-by-construction algorithms, which perform well for hundreds of agents.
D. Panagou, M. Turpin and V. Kumar "Decentralized goal assignment and trajectory generation in multi-robot networks: A multiple Lyapunov functions approach", 2014 IEEE Int. Conf. on Robotics and Automation, Hong Kong, China, June 2014
Visibility Maintenance for Leader-Follower Formations in Obstacle Environments: We consider GPS-denied obstacle environments where multiple robots need to coordinate their motion using vision-based sensing systems only, in the absence of explicit information exchange. Physical obstacles may obstruct visibility, therefore effective sensing, and furthermore should always be avoided.
We develop decentralized motion coordination algorithms for formations of mobile robots in such constrained environments, which guarantee the collision-free motion of the robotic network and the maintenance of visibility among robotic agents.
D. Panagou and V. Kumar "Cooperative visibility maintenance for leader-follower formations in obstacle environments", IEEE Transactions on Robotics, vol. 30, no. 4, pp. 831-844, Aug. 2014
Dynamic Positioning and Formation Control for Underactuated Marine Vehicles: Guidance, navigation and control of marine vehicles (ships, surface vessels and underwater vehicles) is an active research topic, motivated in part by the extensive use of autonomous vehicles in oil industry, scientific explorations (e.g. in oceanographic, archaeological and marine biology research), search and rescue missions, surveillance and inspection tasks, etc. The underwater environment, in particular, poses additional challenges to guidance, navigation and control tasks due to the lack of GPS measurements. Thus, vision is often the main means of sensing and localization with respect to targets of interest.
We develop hybrid and switching control algorithms for underactuated vehicles which move in the presence of unknown external disturbances and vision-based constraints, which guarantee the practical stability of the system with respect to a target of interest. Trade-offs between visibility maintenance and accurate positioning are studied and analyzed. We also consider multi-vehicle formation control so that multiple marine vehicles maintain visibility with, and eventually encircle, a target of interest.
Our next goals include the extension of the methodologies to docking and collaborative control problems for satellites and spacecraft.
D. Panagou and K. J. Kyriakopoulos "Viability control for a class of underactuated systems", Automatica, 49 (2013), pp. 17-29
D. Panagou and K. J. Kyriakopoulos "Dynamic positioning for an underactuated marine vehicle using hybrid control", International Journal of Control, 2013, http://dx.doi.org/10.1080/00207179.2013.828853
D. Panagou and K. J. Kyriakopoulos "Cooperative formation control of underactuated marine vehicles for target surveillance under sensing and communication constraints", 2013 IEEE Int. Conf. on Robotics and Automation, Karlsruhe, Germany, May 2013